Agents you configure, not a black box.
Every AI agent in Caspira Live is yours to shape: the system prompt, the model, the temperature, the reply delay, and exactly which documents it may answer from.
Six settings that decide how your agent behaves.
All of them live in the dashboard. None of them require a developer.
System prompt
Write how the agent should behave, what tone to use, and what it must never promise. This is plain text you own and can revise at any time.
Model selection
Choose the model each agent runs on. Different agents in the same workspace can use different models.
Temperature
Tune how closely replies stick to the source material. Lower values keep answers conservative and predictable.
Reply delay
Set how long the agent waits before answering, leaving your team a window to pick up the conversation first.
Widget binding
Attach a specific agent to your chat widget, so the right persona answers on the right surface.
Knowledge scope
Point the agent at the documents it may answer from. It retrieves the closest passages at reply time rather than improvising.
Answers that come from your own writing.
The agent does not answer from general world knowledge. It retrieves passages from documents your team has added, which means the reply reflects your policies and your wording.
- Add articles, policies, and FAQs as documents
- Caspira splits each document into passages and indexes them
- At reply time the closest passages are retrieved
- The agent drafts an answer grounded in those passages
- Anything outside your content becomes a ticket instead
Quality of answers tracks the quality of your knowledge base. Thin documents produce thin coverage and more escalations.
You can see what the agent did.
Automation is only useful if you can review it afterwards.
AI versus agent replies
Reporting separates messages sent by the AI from messages sent by your team, so you can see how much the agent is genuinely covering.
Manual drafting
Agents can trigger a draft from the inbox and edit it before sending, keeping a human in the loop on sensitive threads.
QA inspection
Conversations can be scored in the inspection workspace, whether they were handled by a person or the AI.
Configure your first agent today.
Every plan, including Free, includes the AI agent and the knowledge base.